3 ms·
> marginal value of each additional one is probably low This is the point of the paper. Specifically, that single embedding vectors are sparse enough that you
by kevmo314 1y ago
> marginal value of each additional one is probably low
This is the point of the paper. Specifically, that single embedding vectors are sparse enough that you can compact more data from additional vectors together to improve retrieval performance.